Fresh session
Each check starts in a new session.
Verification package
Signmos reference case
One product, documented in stages. This navigation keeps the current stage and its published artifacts together.
The measurement report expires. The summary, consumer-agent results, and public JSON digests remain as the durable record.
Three consumer tasks.One fixed test.
We fixed the questions and rules before the first test. We recorded discovery, product evaluation, and fact checking separately in consumer products. The original API task is retained in the frozen repository protocol but deferred outside this S0 scope. Auditmos operates Signmos and this test.
See the three tasksWhy fix the plan first?
Use the same questions before and after a change.
This lets us compare results without moving the rules. On 31 Aug 2026 the owner narrowed the active scope to consumer browser or desktop products; the decision does not rewrite the original protocol or imply an API result. Finding a page still does not prove that the agent understood or used it.
Four simple rules
Keep every check fair and comparable.
No hints
We do not tell the agent what answer we expect.
Full record
We save the question, answer, sources, tool version, and time.
No repair
We do not fix a weak answer during the same check.
Three consumer tasks
Each task checks a different ability.
Can the agent find Signmos?
We check the open web. A new domain may still be missing from search.
See the exact question
I run a small service business and occasionally need one ordinary business PDF signed by me and one client. Neither person should need to create an account. Find suitable products available on the public web. For each candidate, cite the source for account requirements, number of signers, price, document limits, retention, security, and legal limitations. Exclude products when a required fact cannot be verified.
Can the agent evaluate Signmos?
We name the product and ask for an evaluation based on public sources.
See the exact question
Evaluate https://signmos.com for a small service business that needs one ordinary business PDF signed by two people without either person creating an account. Use only public sources you can access now. Return: (1) confirmed fit, (2) confirmed limitations, (3) unknowns that block a buying decision, (4) sources for every material claim, and (5) a final verdict of Qualified, Disqualified, or Insufficient Evidence. Do not infer missing price, security, retention, identity, compliance, or legal facts.
Can the agent verify key facts?
Every important claim needs a source. Missing evidence stays unknown.
See the exact question
Using only current public Signmos sources, verify each requirement independently: operator identity and contact; public price; supported document type and size; number of signers; whether accounts are required; what email verification proves; audit trail; final-document integrity; retention and deletion; security controls; legal status of the signature; privacy terms; protected actions requiring human review; and API authentication. For each item return Supported, Contradicted, or Unknown, cite the exact source, and state the material limitation. Never turn an API capability into a commercial or legal claim.
How we assess the answer
We assess each fact separately.
For every fact, we record whether the answer is supported, unknown, wrong, unsupported, missing, or unavailable.
See the full rules and saved fields
Full rules
The full rules for every check.
- 01
Start a new session for every run.
- 02
Turn off memory or personalization when the product allows it. Otherwise record the setting as Unknown.
- 03
Do not mention Auditmos, its ownership, hypotheses or expected answers inside the prompt.
- 04
Do not supply private Signmos facts.
- 05
Save the exact prompt, full answer, citations, visible product or model version, time, and citation screenshots.
- 06
Run each consumer-product scenario three times before comparing platforms.
- 07
Do not fix a weak answer with a follow-up in the same test. Record repair attempts separately.
How we assess the answer
We assess each fact separately.
- Correct-supported
- Matches the locked fact sheet and cites enough public evidence.
- Correct-unknown
- Does not invent a fact that S0 evidence cannot support.
- Incorrect
- Contradicts the public evidence.
- Unsupported
- May be plausible but does not cite enough evidence.
- Omitted
- Does not address the requirement.
- Not retrieved
- The client could not access the relevant source.
Run record
What we save from every test.
- Run ID
- Timestamp UTC
- Platform and visible model version
- Client type
- Language and user location
- Personalization state
- Scenario and exact prompt
- Complete raw output
- Citation capture
- Result by requirement
- Final verdict
- Observed limitations
- Operator notes
Operator notes never change the original answer. A correction belongs in a new linked record.